Table 6.

Comparison of different deep-learning-based models (%).

Model	BERT-GRU-CRF (%)	BERT-LSTM-CRF (%)	BERT-LSTM-CRF (%)	ALBERT-LSTM-CRF (%)	RoBERTa-GRU-CRF (%)	RoBERTa-LSTM-CRF (%)
Cheating	0	0	35.00	3.60	93.24	89.33
Mixed	95.94	87.66	96.71	50.35	99.39	99.54
Out-listed	84.88	41.90	83.95	74.24	96.60	95.47
Not fresh	0	0	63.72	66.00	89.92	91.06
Frozen	65.10	60.16	69.02	53.11	81.10	81.42
Spoilage	60.44	18.12	54.62	40.72	94.07	87.18
Over-loading	51.03	24.79	86.29	86.30	99.27	98.87
Deep-processing	64.17	7.21	57.10	48.81	82.59	78.83
Mean value	52.70	29.98	68.30	52.89	92.02	90.21